JP2007241765A - Method for acquiring attribute expression, and method, device and program for acquiring evaluation expression - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To acquire even an attribute expression set even when co-occurrence frequency of an attribute expression with an input evaluation expression is low, and to acquire even an evaluation expression set even when co-occurrence frequency of an evaluation expression with an input attribution expression is low. <P>SOLUTION: When the evaluation expression is input, attribution expression candidates are extracted based on information about the appearing position of a word from a document storage means, and the attribution expression candidates are narrowed down utilizing the appearing frequency of the attribution expression candidates. Then, for each narrowed-down attribution expression candidate, co-occurrence words are extracted from the document storage means based on the information about the appearing position of the word. The attribution expression candidates having the same co-occurrence word equal to or more than a predetermined number are collected and extracted as the attribute expression set. The evaluation expression set is extracted similarly. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to an attribute expression acquisition method, an evaluation expression acquisition method, an apparatus, and a program, and in particular, an evaluation expression that represents an evaluation of the attribute expression based on the attribute expression, or what the evaluation expression is based on the evaluation expression. The present invention relates to an attribute expression acquisition method, an evaluation expression acquisition method, an apparatus, and a program for acquiring an attribute expression indicating whether the evaluation is for a document from a document.

When acquiring from a document an evaluation expression that represents an evaluation of the attribute expression based on the attribute expression or an attribute expression that indicates what the evaluation expression is based on the evaluation expression, within the same sentence A method using statistical information such as the frequency at which attribute expressions and evaluation expressions appear simultaneously is known. For example, there is a known method for obtaining an attribute expression (such as “screen” or “design”) indicating the property of an object from an evaluation expression (such as “small” or “beautiful”) regarding the object such as a mobile phone ( For example, refer nonpatent literature 1).
"Collection of evaluation expressions for opinion extraction", Proc. Of Language Processing Society, Vol.12, No.3, pp.203-222, 2005

  However, since the conventional method for obtaining an evaluation expression uses statistical information, it is difficult to automatically obtain a word with a low frequency. For example, when acquiring an attribute expression from an evaluation expression, since there are very many types of expression, there are many attribute expressions in which a specific combination of an evaluation expression and an attribute expression appears less frequently in the same sentence, There is a problem that many attribute expressions with such a low frequency cannot be automatically acquired.

  The present invention has been made in view of the above points, and by examining whether an attribute expression candidate has the same co-occurrence word other than the input evaluation expression, even if the attribute expression is less frequent, Attribute expression acquisition method that can be recognized as an evaluation expression by examining whether the evaluation expression candidates have the same co-occurrence word other than the input attribute expression, even if the evaluation expression can be recognized and less frequently Another object of the present invention is to provide an evaluation expression acquisition method, apparatus, and program.

  FIG. 1 is a diagram (part 1) for explaining the principle of the present invention.

The present invention (Claim 1) includes evaluation expression input means, attribute expression candidate extraction means, attribute expression candidate storage means, attribute expression candidate narrowing means, co-occurrence word extraction means, co-occurrence word storage means, attribute expression extraction means, and a plurality of attribute expression extraction means An attribute expression acquisition method for acquiring, from a document, an attribute expression representing the property of a target object based on an evaluation expression that is an expression representing an evaluation of the target object in an attribute expression acquisition device having a storage means. And
An evaluation expression input step (step 1) in which the evaluation expression input means inputs one evaluation expression;
An attribute expression candidate extraction unit is a document storage unit that stores a word-delimited document prepared in advance as an expression (attribute expression candidate) that simultaneously appears in the same sentence as one evaluation expression acquired by the evaluation expression input step. An attribute expression candidate extraction step (step 2) for extracting based on information on the appearance position of the word;
An attribute expression candidate storage means for storing an attribute expression candidate in the attribute expression candidate storage means for each sentence in an attribute expression candidate storage step (step 3);
An attribute expression candidate narrowing-down means for narrowing down the attribute expression candidates by using the appearance frequency of the attribute expression candidates from the attribute expression candidate storage means (step 4);
For each attribute expression candidate narrowed down by the attribute expression candidate narrowing-down step, the co-occurrence word extracting means selects a word (co-occurrence word) other than the evaluation expression that simultaneously appears in the same sentence as the attribute expression candidate. A co-occurrence word extraction step (step 5) for extracting from the document storage means that stores the document with the word separator based on the information about the position;
A co-occurrence word storage means for storing a set of co-occurrence words for each attribute expression candidate in the co-occurrence word storage means (step 6);
Attribute expression extraction means extracts attribute expression candidates having the same co-occurrence word more than a predetermined number from the set of co-occurrence words stored in the co-occurrence word storage means and extracts them as an attribute expression set An expression extraction step (step 7) is performed.

Further, the present invention (Claim 2), in the attribute expression candidate extraction step (Step 2),
The attribute expression candidate extraction means performs a step of extracting attribute expression candidates using the language information from the document storage means storing the document with language information storing the language information about the word itself of the document prepared in advance,
In the co-occurrence word extraction step (step 5),
The co-occurrence word extraction unit performs a step of extracting the co-occurrence word from the document storage unit storing the document with language information using the language information.

Further, the present invention (Claim 3), in the attribute expression candidate extraction step (Step 2),
The attribute expression candidate extraction means performs a step of extracting attribute expression candidates from the document storage means storing a document with syntax information related to words of a document prepared in advance using the syntax information;
In the co-occurrence word extraction step (step 5),
The co-occurrence word extraction unit performs a step of extracting the co-occurrence word using the syntax information from the document storage unit storing the document with syntax information.

  FIG. 2 is a diagram illustrating the principle of the present invention (part 1).

The present invention (Claim 4) is an attribute expression acquisition device for acquiring, from a document, an attribute expression that represents the property of a target object, based on an evaluation expression that is an expression that represents the evaluation of the target object.
Evaluation expression input means 110 for inputting one evaluation expression;
Expressions (attribute expression candidates) that simultaneously appear in the same sentence as one evaluation expression acquired by the evaluation expression input unit 110 are related to the appearance position of the word from the document storage unit 10 that stores a word-delimited document prepared in advance. Attribute expression candidate extraction means 120 for extracting based on information;
Attribute expression candidate storage means 130 for storing attribute expression candidates in the attribute expression candidate storage means 20 for each sentence;
Attribute expression candidate narrowing means 140 for narrowing down the attribute expression candidates using the appearance frequency of the attribute expression candidates from the attribute expression candidate storage means 20;
For each attribute expression candidate narrowed down by the attribute expression candidate narrowing means 140, a word (co-occurrence word) other than the evaluation expression that simultaneously appears in the same sentence as the attribute expression candidate is determined based on information on the appearance position of the word. A co-occurrence word extracting means 150 for extracting the document with word delimiter from the document storage means 10;
Co-occurrence word storage means 160 for storing a set of co-occurrence words for each attribute expression candidate in the co-occurrence word storage means 30;
An attribute expression extraction unit 170 that collectively extracts attribute expression candidates having the same co-occurrence word from a set of co-occurrence words stored in the co-occurrence word storage unit 30 as an attribute expression set. Have.

Further, according to the present invention (Claim 5), in the attribute expression candidate extraction unit 120,
Including means for extracting attribute expression candidates from the document storage means 10 storing a document with language information relating to the word itself of the document prepared in advance using the language information;
The co-occurrence word extraction means 150
Means for extracting a co-occurrence word using the language information from the document storage means 10 storing the document with language information;

Further, according to the present invention (claim 6), the attribute expression candidate extraction unit 120
Means for extracting attribute expression candidates from the document storage means 10 storing a document with syntax information related to words of a document prepared in advance, using the syntax information;
The co-occurrence word extraction means 150
Means for extracting a co-occurrence word from the document storage means 10 storing the document with syntax information using the syntax information;

  The present invention (Claim 7) is an attribute expression acquisition program for causing a computer to function as the attribute expression acquisition apparatus according to claims 4 to 6.

  FIG. 3 is a diagram (part 2) for explaining the principle of the present invention.

The present invention (Claim 8) includes attribute expression input means, evaluation expression candidate extraction means, evaluation expression candidate storage means, evaluation expression candidate narrowing means, co-occurrence word extraction means, co-occurrence word storage means, evaluation expression extraction means and a plurality of An evaluation expression acquisition method for acquiring, from a document, an evaluation expression, which is an expression representing an evaluation of an object thing, based on an attribute expression representing a property of the object object in an evaluation expression acquisition apparatus having a storage means ,
An attribute expression input step (step 11) in which the attribute expression input means inputs one attribute expression;
An evaluation expression candidate extraction unit uses a word from a document storage unit storing a word-delimited document prepared in advance as an expression (evaluation expression candidate) that simultaneously appears in the same sentence as one attribute expression acquired in the attribute expression input step. An evaluation expression candidate extraction step (step 12) for extracting based on information on the appearance position of
An evaluation expression candidate storage unit (step 13) in which the evaluation expression candidate storage unit stores the evaluation expression candidate in the evaluation expression candidate storage unit for each sentence;
An evaluation expression candidate narrowing-down step (step 14) for narrowing down the evaluation expression candidates from the evaluation expression candidate storage means using the appearance frequency of the evaluation expression candidates from the evaluation expression candidate storage means;
For each evaluation expression candidate narrowed down by the evaluation expression candidate narrowing step (step 14), the co-occurrence word extracting means selects a word (co-occurrence word) other than the attribute expression that simultaneously appears in the same sentence as the evaluation expression candidate. A co-occurrence word extraction step (step 15) for extracting from the document storage means storing the document with word delimiters based on the information on the appearance position of the word;
A co-occurrence word storage means for storing a set of co-occurrence words in the co-occurrence word storage means for each evaluation expression candidate (step 16);
Evaluation in which the evaluation expression extraction unit collectively extracts evaluation expression candidates having the same co-occurrence word in a predetermined number or more from the set of co-occurrence words stored in the co-occurrence word storage unit as an evaluation expression set An expression extraction step (step 17) is performed.

Further, the present invention (Claim 9), in the evaluation expression candidate extraction step (Step 12),
The evaluation expression candidate extraction means performs a step of extracting evaluation expression candidates using the language information from the document storage means storing the document with language information storing the language information about the word itself of the prepared document,
In the co-occurrence word extraction step (step 15),
The co-occurrence word extraction unit performs a step of extracting the co-occurrence word using the language information from the document storage unit storing the document with language information.

Further, the present invention (Claim 10), in the evaluation expression candidate extraction step (Step 12),
The evaluation expression candidate extraction means performs a step of extracting evaluation expression candidates using the syntax information from a document storage means storing a document with syntax information related to words of a document prepared in advance,
In the co-occurrence word extraction step (step 15),
The co-occurrence word extraction unit performs a step of extracting the co-occurrence word from the document storage unit storing the document with syntax information using the syntax information.

  FIG. 4 is a principle configuration diagram (part 2) of the present invention.

The present invention (Claim 11) is an evaluation expression acquisition device for acquiring an evaluation expression, which is an expression representing an evaluation of a target thing, from a document based on an attribute expression indicating the property of the target thing.
Attribute expression input means 210 for inputting one attribute expression;
Expressions (evaluation expression candidates) that simultaneously appear in the same sentence as one attribute expression acquired by the attribute expression input unit 210 are related to the appearance position of the word from the document storage unit 40 that stores a word-delimited document prepared in advance. Evaluation expression candidate extraction means 220 for extracting based on the information;
Evaluation expression candidate storage means 230 for storing evaluation expression candidates in the evaluation expression candidate storage means 50 for each sentence;
An evaluation expression candidate narrowing means 240 that narrows down evaluation expression candidates from the evaluation expression candidate storage means 50 by using the appearance frequency of evaluation expression candidates;
For each evaluation expression candidate narrowed down by the evaluation expression candidate narrowing means 240, a word (co-occurrence word) other than the attribute expression that simultaneously appears in the same sentence as the evaluation expression candidate is determined based on information on the appearance position of the word. A co-occurrence word extracting means 250 for extracting the document with word delimiters from the document storage means 40;
Co-occurrence word storage means 260 for storing a set of co-occurrence words for each evaluation expression candidate in the co-occurrence word storage means 60;
An evaluation expression extraction unit 270 that collectively extracts evaluation expression candidates having the same co-occurrence word from a set of co-occurrence words stored in the co-occurrence word storage unit 60 as an evaluation expression set. Have.

Further, according to the present invention (Claim 12), in the evaluation expression candidate extraction unit 220,
Including means for extracting evaluation expression candidates using the language information from the document storage means 40 storing a document with language information relating to the word itself of the document prepared in advance,
The co-occurrence word extraction means 250
Means for extracting co-occurrence words using the language information from the document storage means 40 storing the document with language information;

The present invention (Claim 13) is characterized in that in the evaluation expression candidate extraction means 220,
Means for extracting evaluation expression candidates from the document storage means 40 storing a document with syntax information related to words of a document prepared in advance, using the syntax information;
The co-occurrence word extraction means 250
Means for extracting a co-occurrence word using the syntax information from the document storage means 40 storing the document with syntax information;

  The present invention (Claim 14) is an evaluation expression acquisition program that causes a computer to function as the evaluation expression acquisition device according to Claims 11 to 13.

  As described above, according to the present invention, an attribute expression having a low co-occurrence frequency with the input evaluation expression is obtained by examining whether or not the attribute expression candidate has the same co-occurrence word other than the input evaluation expression. It is also possible to obtain an evaluation expression that has a low co-occurrence frequency with the input attribute expression by examining whether the evaluation expression candidate has the same co-occurrence word other than the input attribute expression. However, there is an effect that it can be acquired.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[First Embodiment]
An attribute expression acquisition device will be described as a first embodiment of the present invention.

  FIG. 5 is a basic block configuration diagram of the attribute expression acquisition device according to the first exemplary embodiment of the present invention.

  The attribute expression acquisition apparatus 100 includes an evaluation expression input unit 110, an attribute expression candidate extraction unit 120, an attribute expression candidate storage unit 130, an attribute expression candidate narrowing unit 140, a co-occurrence word extraction unit 150, a co-occurrence word storage unit 160, an attribute expression The extraction unit 170, the language information / syntax information / document separator-attached document storage unit 10, the attribute expression candidate storage unit 20, and the co-occurrence word storage unit 30 are configured.

  The language information / syntax information / word-delimited document storage unit 10, the attribute expression candidate storage unit 20, and the co-occurrence word storage unit 30 are storage devices such as a disk device and a memory.

  The “attribute expression” used below is, for example, an input evaluation expression within the same sentence if the input expression is an evaluation expression (such as “broad” or “beautiful”) related to an object such as a mobile phone. This refers to attribute expressions (such as “area” and “screen”) that represent the properties of the target object that appear as a pair relationship.

  The evaluation expression input unit 110 inputs one evaluation expression.

  The attribute expression candidate extraction unit 120 prepares language information / syntax information / word-delimited document prepared in advance for expressions (attribute expression candidates) that simultaneously appear in the same sentence as one evaluation expression acquired by the evaluation expression input unit 110. Extracted from the storage unit 10 based on information on words such as parts of speech and information on syntax such as dependency.

  The attribute expression candidate storage unit 130 stores the attribute expression candidates in the attribute expression candidate storage unit 20 for each sentence.

  The attribute expression candidate narrowing-down unit 140 narrows down the attribute expression candidates from the attribute expression candidate storage unit 20 by using the appearance frequency of the attribute expression candidates.

  The co-occurrence word extraction unit 150 prepares in advance words (co-occurrence words) other than the input evaluation expression that simultaneously appear in the same sentence as the attribute expression candidates for each attribute expression candidate narrowed down by the attribute expression candidate narrowing unit 140. The extracted language information / syntax information / word-delimited document storage unit 10 is extracted based on information on words such as parts of speech and information on syntax such as dependency.

  The co-occurrence word storage unit 160 stores a set of co-occurrence words in the co-occurrence word storage unit 30 for each attribute expression candidate.

  The attribute expression extraction unit 170 collectively extracts attribute expression candidates having the same number of co-occurrence words from a set of co-occurrence words stored in the co-occurrence word storage unit 30 as an attribute expression set. .

  Next, the operation of the attribute expression acquisition device will be described.

  FIG. 6 is a flowchart of attribute expression acquisition processing according to the first embodiment of this invention.

  An example of acquiring an attribute expression when an evaluation expression “many” is input will be described based on the flowchart of FIG.

  FIG. 7 is an example of a sentence for each sentence ID in the present embodiment, and FIGS. 8 to 10 show examples of contents of language information, syntax information, and word-delimited document storage unit 10.

  Step 101) One evaluation expression is input by the evaluation expression input unit 110.

  In this example, “many” is input. Although one evaluation expression is input here, a method may be used in which a set of evaluation expressions input in advance is stored in a storage medium such as a memory or a disk, and evaluation expressions are input one by one from there.

  Step 102) The linguistic information, syntax information, and words prepared in advance by the attribute expression candidate extraction unit 120 include expressions (attribute expression candidates) that simultaneously appear in the same sentence as one evaluation expression acquired by the evaluation expression input unit 10. A word having a part of speech as a noun and having a subject predicate relationship (subject relationship) is extracted from the delimited document storage unit 10.

  For example, for the sentence shown in FIG. 7, words are separated using a generally known morphological analyzer, parts of speech are extracted as linguistic information, and main sentences are described as syntactic information using a generally known parser. FIG. 8 shows an example in which the relationship retrieval result is stored in the document storage unit 10 with language information, syntax information, and word delimiters. The column “(main)” of the main predicate relationship in FIG. 8 indicates a word that has a relationship between the word and the subject, and “(predicate)” indicates a word that has a relationship between the word and the predicate.

  For example, since the predicate of “screen” of the sentence ID 1 “many beautiful screens” is “many”, the main description relation column of “screen” is “(description) many”. On the other hand, since the subject of “many” in the same sentence is “screen”, the column of “main” predicate relationship is described as “(main) screen”. Here, the explanation is made by using the part of speech as the language information and the predicate relationship as the syntax information. However, the present invention is obtained by a morphological analyzer and a syntax analyzer and is effective in the present invention (for example, tense). , Utilization, etc.) and syntax information (for example, taking out a main word and a main sentence) is not limited to this. In this example, a word that is a noun and has a main description relationship with the input evaluation expression “many” is extracted as an attribute expression candidate. For example, in the case of sentence ID 1 “many beautiful screens”, “screen” Since the word is a noun and has a main predicate relationship with the input evaluation expression “many”, it is output as an attribute expression candidate. In this example, there is only one attribute expression candidate for each sentence, but a plurality of attribute expression candidates can be extracted from one sentence if the extraction condition is met.

  Step 103) The attribute expression candidate storage unit 130 stores the attribute expression candidates extracted in step 102 in the attribute expression candidate storage unit 20 for each sentence.

  FIG. 11 shows the result of attribute expression candidates extracted when the input evaluation expression is “many”. "-" In the figure indicates that no attribute expression candidate was extracted from the sentence.

  Step 104) The attribute expression candidate narrowing-down unit 140 narrows down the attribute expression candidates that appear only once from the attribute expression candidate storage unit 20. Here, the information is once, but the present invention is not limited to this. For example, any information related to the frequency of attribute expression candidates may be used, such as referring to the appearance distribution of the entire attribute expression candidates. In addition, a frequently-used item may be recognized as an attribute expression at this time and taken out. For example, when it is determined that the frequency of twice or more is highly reliable as the attribute expression, “number of times” is recognized as the attribute expression as it is from FIG. In this case, the attribute expression is added to the attribute expression set regardless of the attribute expression extraction conditions in step 107 described later. In step 104, in this example, the attribute expression candidates are narrowed down to “double-sided”, “area”, and “noise” from FIG. 11 stored in the attribute expression candidate storage unit 20.

  Step 105) The co-occurrence word extraction unit 150 previously prepares words (co-occurrence words) other than the input evaluation expression that appear simultaneously in the same sentence as the attribute expression candidates for each attribute expression candidate acquired in Step 104. A word having a part of speech as an adjective and having a subject predicate relationship (subject relationship) is extracted from the language information, syntax information, and word-delimited document storage unit 10.

  Here, the explanation is made using the part of speech as the language information and the predicate relationship as the syntax information. , Utilization, etc.) and syntactic information (for example, taking out main words and main sentences), the present invention is not limited to this. Also, it may be different from the information used in step 102 above.

  In this example, words that are adjectives and have a predicate relationship with the attribute expression candidates extracted in step 104 and are not input as many evaluation expressions are extracted as co-occurrence words. In the case of the sentence ID 6 “the screen of this model is beautiful”, “beautiful” is an adjective and is a word that has a predicate relationship with the attribute expression candidate “screen” and is different from the input evaluation expression “many” So it is extracted as a co-occurrence word. The co-occurrence words shown in FIG. 12 are extracted from the examples of the language information, syntax information, and word-delimited document storage unit 10 shown in FIGS. In this example, only one is extracted per sentence, but a plurality of co-occurrence words can be extracted if the extraction conditions are met.

  Step 106) The co-occurrence word storage unit 160 stores a set of co-occurrence words in the co-occurrence word storage unit 30 for each attribute expression candidate. In the case of this example, the co-occurrence words extracted in step 105 are stored in the co-occurrence word storage unit 30 for each attribute expression candidate as shown in FIG.

  Step 107) The attribute expression extraction unit 170 collectively extracts one or more attribute expression candidates having the same co-occurrence word from the set of co-occurrence words stored in the co-occurrence word storage unit 30 as an attribute expression set. .

  In this example, from the co-occurrence set of FIG. 13 stored in the co-occurrence word storage unit 30, at least one same co-occurrence word “wide” is stored for the attribute expression candidates “screen” and “area”. Therefore, “screen” and “area” are extracted as the attribute expression set. “Noise” is not extracted because it does not meet the extraction conditions. In this example, the number of the same co-occurrence words is one or more. However, any integer can be specified as long as it is an integer. In addition, the words already recognized as attribute expressions in step 104 can be unconditionally added to the attribute expression set extracted in step 107. For example, if an attribute expression candidate that appears twice or more is recognized as an attribute expression in step 104, “number of times” is added to the attribute expression set in the example of FIG.

  In this way, even if the combination of the input evaluation expression and the attribute expression candidate appears only infrequently in the same sentence, check whether the attribute expression candidate has the same co-occurrence word other than the input evaluation expression. It can be recognized as an attribute expression, and an attribute expression set can be acquired.

[Second Embodiment]
Hereinafter, an evaluation expression acquisition apparatus will be described as a second embodiment of the present invention.

  FIG. 14 is a basic block configuration diagram of the evaluation expression acquiring apparatus according to the second embodiment of the present invention.

  The evaluation expression acquisition apparatus 200 includes an attribute expression input unit 210, an evaluation expression candidate extraction unit 220, an evaluation expression candidate storage unit 230, an evaluation expression candidate narrowing unit 240, a co-occurrence word extraction unit 250, a co-occurrence word storage unit 260, an evaluation expression. It comprises an extraction unit 270, language information / syntax information / document-with-word-breaking document storage unit 40, evaluation expression candidate storage unit 50, and co-occurrence word storage unit 60.

  The language information / syntax information / word-delimited document storage unit 40, the evaluation expression candidate storage unit 50, and the co-occurrence word storage unit 60 are storage media such as a disk device and a memory.

  The attribute expression input unit 210 inputs one attribute expression.

  The evaluation expression candidate extraction unit 220 obtains expressions (evaluation expression candidates) that simultaneously appear in the same sentence as one attribute expression from words, parts of speech, and the like from the language information / syntax information / word delimited document storage unit 40 prepared in advance. Information based on information on the syntax and information on the dependency, etc.

  The evaluation expression candidate storage unit 230 stores evaluation expression candidates in the evaluation expression candidate storage unit 50 for each sentence.

  The evaluation expression candidate narrowing unit 240 uses the appearance frequency of evaluation expression candidates from the evaluation expression candidate storage unit 50 to narrow down evaluation expression candidates.

  The co-occurrence word extraction unit 250 prepares in advance words (co-occurrence words) other than the input attribute expression that simultaneously appear in the same sentence as the evaluation expression candidate for each evaluation expression candidate narrowed down by the evaluation expression candidate narrowing unit 240. The extracted language information / syntax information / word-delimited document storage unit 40 is extracted based on information on words such as parts of speech and information on syntax such as dependency.

  The co-occurrence word storage unit 260 stores a set of co-occurrence words in the co-occurrence word storage unit 60 for each evaluation expression candidate.

  The evaluation expression extraction unit 270 collectively extracts evaluation expression candidates having the same number of co-occurrence words from a set of co-occurrence words stored in the co-occurrence word storage unit 60 as an evaluation expression set. .

  Next, the operation of the evaluation expression acquisition apparatus 200 will be described.

  FIG. 15 is a flowchart of evaluation expression acquisition processing according to the second embodiment of this invention. An example of obtaining an evaluation expression when an attribute expression “operation” is input will be described based on the flowchart of FIG.

  Step 201) The attribute expression input unit 210 inputs one attribute expression.

  In this example, “operation” is input. Here, one attribute expression is input, but a method may be used in which a set of attribute expressions to be input in advance is stored in a storage medium such as a memory or a disk, and the attribute expressions are input one by one from there.

  Step 202) Language expression / syntax information / word prepared in advance by the evaluation expression candidate extraction unit 220 as expressions (evaluation expression candidates) that simultaneously appear in the same sentence as one attribute expression acquired by the attribute expression input unit 210 A word having a part of speech as an adjective and having a subject predicate relationship (subject relationship) is extracted from the delimited document storage unit 40. For example, with respect to the sentence shown in FIG. 16, words are separated using generally known morphological analyzers and syntactic analyzers, the part of speech as linguistic information, and the result of extracting the main predicate relationship as linguistic information, FIG. 17 to FIG. 19 show examples stored in the document storage unit 40 with word breaks. In the document storage unit 40 with language information / syntax information / word delimiter, as in FIGS. 8 to 10, the column “(main)” in FIGS. 17 to 19 has a relationship between the word and the subject. “(Predicate)” indicates a word in the relationship of the predicate of the word. For example, since the predicate of “motion” of the sentence ID 1 “focus action is fast” is “fast”, the column of the main action relationship of “action” is “(description) fast”. On the other hand, since the subject of “fast” in the same sentence is “motion”, the “fast” main predicate relationship column is described as “(main) motion”.

  Here, the explanation is made by using the part of speech as the language information and the predicate relationship as the syntax information. However, the present invention is obtained by a morphological analyzer and a syntax analyzer and is effective in the present invention (for example, tense). , Utilization, etc.) and syntax information (for example, taking out a main word and a main sentence) is not limited to this. In this example, an adjective and a word that has a main predicative relationship with the input attribute expression “motion” are extracted as evaluation expression candidates. For example, in the case of sentence ID 1 “the focus operation is fast”, “fast” Is an adjective and is a word that is in a predicate relationship with the input attribute expression “action”, and is extracted as an evaluation expression candidate. In this example, there is only one evaluation expression candidate for each sentence, but a plurality of evaluation expression candidates can be extracted from one sentence if the extraction condition is met.

  Step 203) The evaluation expression candidate storage unit 230 stores the evaluation expression candidates extracted in step 202 in the evaluation expression candidate storage unit 50 for each sentence. FIG. 20 shows the result of the evaluation expression candidate extracted when the input attribute expression is “motion”. “-” In FIG. 20 indicates that no evaluation expression candidate is extracted from the sentence.

  Step 204) The evaluation expression candidate narrowing-down unit 240 narrows down the evaluation expression candidates from the evaluation expression candidate storage unit 50 to evaluation expression candidates that appear only once. Here, it is assumed to be once, but the present invention is not limited to this. For example, any information related to the frequency of evaluation expression candidates may be used, such as referring to the appearance distribution of all evaluation expression candidates. Moreover, what has a high frequency may be taken out as an evaluation expression as it is at this point. For example, when it is determined that the frequency of two or more times is highly reliable as the evaluation expression, “fast” is recognized as the evaluation expression as it is from FIG. In this case, the evaluation expression is added to the evaluation expression set regardless of the evaluation expression extraction condition in step 207 described later. In step 204, in this example, the evaluation expression candidates are narrowed down to “dull”, “slow”, and “heavy” from the contents of FIG. 20 stored in the evaluation expression candidate storage unit 50.

  Step 205) The co-occurrence word extraction unit 260 prepares in advance words (co-occurrence words) other than the input attribute expression that simultaneously appear in the same sentence as the evaluation expression candidate for each evaluation expression candidate acquired in Step 204. A word having a part of speech as a noun and having a subject expression predicate relationship (subject relationship) is extracted from the language information / syntax information / word delimited document storage unit 40.

  Here, the explanation is made by using the part of speech as the language information and the predicate relationship as the syntax information. However, the present invention is obtained by a morphological analyzer and a syntax analyzer and is effective in the present invention (for example, tense). , Utilization, etc.) and syntactic information (for example, taking out main words and main sentences), the present invention is not limited to this. Further, it may be different from the information used in step 202. In this example, a word that is a noun and has a main-state relationship with the evaluation expression candidate extracted in step 204 and is not the input attribute expression “action” is extracted as a co-occurrence word. If the button response is "dull", "reaction" is a noun and is a word that has a predicate relationship with the evaluation expression candidate "dull" and is different from the input attribute expression "action". Extract as The co-occurrence words shown in FIG. 21 are extracted from the examples of FIGS. In this example, only one is extracted per sentence, but a plurality of co-occurrence words can be extracted if the extraction conditions are met.

  Step 206) The co-occurrence word storage unit 260 stores a set of co-occurrence words in the co-occurrence word storage unit 60 for each evaluation expression candidate. In the case of this example, for the co-occurrence words extracted in step 205, a set of co-occurrence words is stored in the co-occurrence word storage unit 60 for each evaluation expression candidate as shown in FIG.

  Step 207) The evaluation expression extraction unit 270 collectively extracts one or more evaluation expression candidates having the same co-occurrence word from the co-occurrence word sets stored in the co-occurrence word storage unit 60 as an evaluation expression set. In this example, from the contents shown in FIG. 22 stored in the co-occurrence word storage unit 60, at least one same co-occurrence word “reaction” is stored for the evaluation expression candidates “dull” and “slow”. Therefore, “dull” and “slow” are extracted as the evaluation expression set. “Heavy” is not extracted because it does not meet the extraction conditions. In this example, the number of the same co-occurrence words is one or more. However, the number can be specified as long as it is an integer. In addition, words that have already been recognized as evaluation expressions in step 204 can be unconditionally added to the evaluation expression set extracted in step 207. For example, if an evaluation expression candidate that appears twice or more in step 204 is recognized as an evaluation expression, “fast” is added to the evaluation expression set in the example of FIG.

  In this way, even if the combination of the input attribute expression and the evaluation expression candidate appears only infrequently in the same sentence, it is checked whether the evaluation expression candidate has the same co-occurrence word other than the input attribute expression. It can be recognized as an evaluation expression and an evaluation expression set can be obtained.

  The functions of the first embodiment and the second embodiment described above may be realized as a program.

  That is, in the first embodiment described above, as a program for acquiring attribute expressions, an evaluation expression input procedure for inputting one evaluation expression, and one evaluation expression acquired by the evaluation expression input procedure in the same sentence Attribute expression candidate extraction procedure for extracting words (attribute expression candidates) that appear at the same time from a document storage unit storing a word-delimited document prepared in advance based on information on the appearance position of the word, and the attribute expression candidates as sentences Attribute expression candidate storage procedure for storing each attribute expression candidate in the attribute expression candidate storage unit, attribute expression candidate narrowing procedure for narrowing down attribute expression candidates using the frequency of appearance of attribute expression candidates from the attribute expression candidate storage unit, and the attribute expression candidates For each attribute expression candidate narrowed down by the narrowing-down procedure, a word (co-occurrence word) other than the input evaluation expression that appears simultaneously in the same sentence as the attribute expression candidate A co-occurrence word extraction procedure for extracting from a document storage unit storing a word-delimited document prepared in advance based on information on the appearance position, and a set of co-occurrence words for each attribute expression candidate are stored in the co-occurrence word storage unit From the co-occurrence word storage procedure and a set of co-occurrence words stored in the co-occurrence word storage unit, attribute expression candidates having the same co-occurrence word for a predetermined number or more are collectively extracted as an attribute expression set. This can be realized by causing the computer to execute the attribute expression extraction procedure.

  The attribute expression candidate extraction procedure includes a procedure of extracting attribute expression candidates from a document storage unit storing language information related to a word of a document prepared in advance using the language information, and the co-occurrence word A program including a procedure for extracting co-occurrence words using the language information from a document storage unit prepared in advance for the extraction procedure may be used.

  The attribute expression candidate extraction procedure includes a procedure for extracting attribute expression candidates from a document storage unit storing syntax information related to words of a document prepared in advance using the syntax information. The word extraction procedure may be a program including a procedure for extracting a co-occurrence word from a document storage unit storing syntax information related to words in a document prepared in advance using the syntax information.

  In addition, as a program for acquiring an evaluation expression in the second embodiment, an attribute expression input procedure for inputting one attribute expression and an attribute expression acquired by the attribute expression input procedure in the same sentence An evaluation expression candidate extraction procedure for extracting simultaneously appearing words (evaluation expression candidates) from a document storage unit storing a word-delimited document prepared in advance based on information on the appearance position of the word, and the evaluation expression candidate storage unit Evaluation expression candidate storing procedure to be stored, evaluation expression candidate narrowing procedure for narrowing down evaluation expression candidates using the appearance frequency of evaluation expression candidates from the evaluation expression candidate storage unit, and evaluation narrowed down by the evaluation expression candidate narrowing procedure For each expression candidate, a word (co-occurrence word) other than the input attribute expression that appears simultaneously in the same sentence as the evaluation expression candidate is related to the appearance position of the word. A co-occurrence word extraction procedure for extracting from a document storage medium with word breaks prepared in advance, and a co-occurrence word storage procedure for storing a set of co-occurrence words for each evaluation expression candidate in the co-occurrence word storage unit; And an evaluation expression extraction procedure for extracting evaluation expression candidates having the same co-occurrence word more than a predetermined number from the set of co-occurrence words stored in the co-occurrence word storage unit as an evaluation expression set. This is realized by causing the computer to execute.

  Further, the evaluation expression candidate extraction procedure includes a procedure for extracting evaluation expression candidates from a document storage unit storing language information about a word of a document prepared in advance using the language information. The word extraction procedure may be a program including a procedure for extracting co-occurrence words from a document storage unit in which language information about a word of a document prepared in advance is stored.

  The evaluation expression candidate extraction procedure includes a procedure for extracting evaluation expression candidates from a document storage unit storing syntax information related to words of a document prepared in advance, using the syntax information. The word extraction procedure may be a program including a procedure for extracting a co-occurrence word from a document storage unit storing syntax information related to words of a document prepared in advance using the syntax information.

  Further, the above program can be stored in a portable storage medium such as a hard disk, a flexible disk, or a CD-ROM, and can be installed or distributed in a computer.

  The present invention is not limited to the above-described embodiment, and various modifications and applications can be made within the scope of the claims.

  The present invention can be applied to a technique for extracting an attribute expression / evaluation expression of an object from a document.

It is FIG. (1) for demonstrating the principle of this invention. It is a principle block diagram (the 1) of this invention. It is FIG. (2) for demonstrating the principle of this invention. It is a principle block diagram (the 2) of this invention. It is a basic block block diagram of the attribute expression acquisition apparatus in the 1st Embodiment of this invention. It is a flowchart of the attribute expression acquisition process in the 1st Embodiment of this invention. It is an example of the sentence in the 1st Embodiment of this invention. It is an example (the 1) of the document storage part with language information, syntax information, and a word delimiter in the 1st Embodiment of this invention. It is an example (the 2) of the document storage part with language information, syntax information, and a word delimiter in the 1st Embodiment of this invention. It is an example (the 3) of the document storage part with language information, syntax information, and a word delimiter in the 1st Embodiment of this invention. It is an example of the attribute expression candidate in the 1st Embodiment of this invention. It is an example of the co-occurrence word in the 1st Embodiment of this invention. It is an example of the co-occurrence word set for every attribute expression candidate in the 1st Embodiment of this invention. It is a basic block block diagram of the evaluation expression acquisition apparatus in the 2nd Embodiment of this invention. It is a flowchart of the evaluation expression acquisition process in the 2nd Embodiment of this invention. It is an example of the sentence in the 2nd Embodiment of this invention. It is an example (the 1) of the document storage part with language information, syntax information, and a word delimiter in the 2nd Embodiment of this invention. It is an example (the 2) of the document storage part with language information, syntax information, and a word delimiter in the 2nd Embodiment of this invention. It is an example (the 3) of the document memory | storage part with language information, syntax information, and a word delimiter in the 2nd Embodiment of this invention. It is an example of the evaluation expression candidate in the 2nd Embodiment of this invention. It is an example of the co-occurrence word in the 2nd Embodiment of this invention. It is an example of the co-occurrence word set in the 2nd Embodiment of this invention.

Explanation of symbols

10. Document storage means, linguistic information / syntax information / document delimited document storage section 20 attribute expression candidate storage means, attribute expression candidate storage section 30 co-occurrence word storage means, co-occurrence word storage section 40 document storage means, language information / syntax Document storage unit with information / word delimiter 50 Evaluation expression candidate storage unit, evaluation expression candidate storage unit 60 Co-occurrence word storage unit, co-occurrence word storage unit 100 Attribute expression acquisition unit 110 Evaluation expression input unit, evaluation expression input unit 120 Attribute expression Candidate extraction means, attribute expression candidate extraction section 130 Attribute expression candidate storage means, attribute expression candidate storage section 140 Attribute expression candidate narrowing means, attribute expression candidate narrowing section 150 Co-occurrence word extraction means, co-occurrence word extraction section 160 Co-occurrence word storage Means, co-occurrence word storage unit 170 attribute expression extraction unit, attribute expression extraction unit 200 evaluation expression acquisition device 210 attribute expression input unit, attribute expression input unit 220 Value expression candidate extraction means, evaluation expression candidate extraction section 230 Evaluation expression candidate storage means, evaluation expression candidate storage section 240 Evaluation expression candidate narrowing means, evaluation expression candidate narrowing section 250 Co-occurrence word extraction means, Co-occurrence word extraction section 260 Co-occurrence Word storage means, co-occurrence word storage section 270 Evaluation expression extraction means, evaluation expression extraction section

Claims (14)

Attribute expression acquisition means having evaluation expression input means, attribute expression candidate extraction means, attribute expression candidate storage means, attribute expression candidate narrowing means, co-occurrence word extraction means, co-occurrence word storage means, attribute expression extraction means, and a plurality of storage means An attribute expression acquisition method for acquiring, from a document, an attribute expression that represents a property of a target object based on an evaluation expression that is an expression that represents an evaluation of the target object in the device,
An evaluation expression input step in which the evaluation expression input means inputs one evaluation expression;
Document storage in which the attribute expression candidate extraction unit stores a word-delimited document prepared in advance for expressions (attribute expression candidates) that simultaneously appear in the same sentence as the one evaluation expression acquired in the evaluation expression input step Means for extracting an attribute expression candidate based on information on the appearance position of the word;
The attribute expression candidate storage means stores the attribute expression candidates in the attribute expression candidate storage means for each sentence, and an attribute expression candidate storage step;
The attribute expression candidate narrowing-down means uses the appearance frequency of the attribute expression candidates from the attribute expression candidate storage means to narrow down the attribute expression candidates, and
For each attribute expression candidate narrowed down by the attribute expression candidate narrowing-down step, the co-occurrence word extracting means selects a word (co-occurrence word) other than the evaluation expression that simultaneously appears in the same sentence as the attribute expression candidate. A co-occurrence word extraction step for extracting from the document storage means that stores the document with the word separator based on the information on the appearance position of
The co-occurrence word storage means stores a set of the co-occurrence words in the co-occurrence word storage means for each attribute expression candidate; and
The attribute expression extraction unit collects attribute expression candidates having the same co-occurrence word more than a predetermined number from the set of co-occurrence words stored in the co-occurrence word storage unit as an attribute expression set. An attribute expression extraction step to extract;
Attribute expression acquisition method characterized by performing In the attribute expression candidate extraction step,
The attribute expression candidate extraction unit performs a step of extracting the attribute expression candidate from the document storage unit storing a document with language information storing language information about a word of the document prepared in advance using the language information. ,
In the co-occurrence word extraction step,
The co-occurrence word extraction means performs a step of extracting the co-occurrence word from the document storage means storing the document with language information using the language information.
The attribute expression acquisition method according to claim 1. In the attribute expression candidate extraction step,
The attribute expression candidate extraction unit performs a step of extracting the attribute expression candidate from the document storage unit storing a document with syntax information related to words of a document prepared in advance using the syntax information;
In the co-occurrence word extraction step,
The attribute expression acquisition method according to claim 1 or 2, wherein the co-occurrence word extraction unit performs the step of extracting the co-occurrence word using the syntax information from a document storage unit storing the document with syntax information. An attribute expression acquisition device for acquiring, from a document, an attribute expression that represents a property of the target object, based on an evaluation expression that is an expression indicating evaluation of the target object,
An evaluation expression input means for inputting one evaluation expression;
Expressions (attribute expression candidates) that appear simultaneously in the same sentence as the one evaluation expression acquired by the evaluation expression input unit are related to the appearance position of the word from the document storage unit that stores the word-delimited document prepared in advance. Attribute expression candidate extraction means for extracting based on information;
Attribute expression candidate storage means for storing the attribute expression candidates in the attribute expression candidate storage means for each sentence;
Attribute expression candidate narrowing means for narrowing down the attribute expression candidates using the appearance frequency of the attribute expression candidates from the attribute expression candidate storage means;
For each attribute expression candidate narrowed down by the attribute expression candidate narrowing means, a word (co-occurrence word) other than the evaluation expression that appears at the same time in the same sentence as the attribute expression candidate is determined based on information on the appearance position of the word. , Co-occurrence word extraction means for extracting from the document storage means storing the word-delimited document;
Co-occurrence word storage means for storing the set of co-occurrence words in the co-occurrence word storage means for each attribute expression candidate;
Attribute expression extraction means for collectively extracting attribute expression candidates having the same co-occurrence word in a predetermined number or more from the set of co-occurrence words stored in the co-occurrence word storage means as an attribute expression set; ,
An attribute expression acquisition device characterized by comprising: The attribute expression candidate extraction means includes:
Means for extracting the attribute expression candidate from the document storage means storing a document with language information relating to the word itself of the document prepared in advance, using the language information;
The co-occurrence word extracting means includes
5. The attribute expression acquisition apparatus according to claim 4, further comprising means for extracting the co-occurrence word using the language information from a document storage means storing the document with language information. The attribute expression candidate extraction means includes:
Means for extracting the attribute expression candidate from the document storage means storing the document with syntax information related to the words of the document prepared in advance, using the syntax information;
The co-occurrence word extracting means includes
6. The attribute expression acquisition apparatus according to claim 4, further comprising means for extracting the co-occurrence word using the syntax information from a document storage means storing the document with syntax information. Computer
An attribute expression acquisition program that functions as the attribute expression acquisition device according to claim 4. Evaluation expression acquisition having attribute expression input means, evaluation expression candidate extraction means, evaluation expression candidate storage means, evaluation expression candidate narrowing means, co-occurrence word extraction means, co-occurrence word storage means, evaluation expression extraction means, and a plurality of storage means An evaluation expression acquisition method for acquiring, from a document, an evaluation expression, which is an expression representing an evaluation of an object thing, based on an attribute expression representing a property of the object object in the apparatus,
An attribute expression input step in which the attribute expression input means inputs one attribute expression;
Document storage in which the evaluation expression candidate extraction unit stores a word-delimited document prepared in advance for an expression (evaluation expression candidate) that simultaneously appears in the same sentence as the one attribute expression acquired in the attribute expression input step. An evaluation expression candidate extraction step for extracting from the information on the appearance position of the word from the means;
The evaluation expression candidate storage means for storing the evaluation expression candidates in the evaluation expression candidate storage means for each sentence;
The evaluation expression candidate narrowing-down means uses the appearance frequency of the evaluation expression candidates from the evaluation expression candidate storage means to narrow down the evaluation expression candidates, and
For each evaluation expression candidate narrowed down by the evaluation expression candidate narrowing-down step, the co-occurrence word extracting means calculates a word (co-occurrence word) other than the attribute expression that simultaneously appears in the same sentence as the evaluation expression candidate. A co-occurrence word extraction step for extracting from the document storage means storing the word-delimited document based on the information on the appearance position;
The co-occurrence word storage means stores the set of co-occurrence words in the co-occurrence word storage means for each evaluation expression candidate;
The evaluation expression extraction means collectively sets evaluation expression candidates having the same co-occurrence words in a predetermined number or more out of the set of co-occurrence words stored in the co-occurrence word storage means as an evaluation expression set. An evaluation expression extraction step to be extracted;
The evaluation expression acquisition method characterized by performing. In the evaluation expression candidate extraction step,
The evaluation expression candidate extraction unit performs a step of extracting the evaluation expression candidate using the language information from a document storage unit that stores a document with language information that stores language information about a word of the document prepared in advance.
In the co-occurrence word extraction step,
The co-occurrence word extracting means performs a step of extracting a co-occurrence word using the language information from a document storage means storing the document with language information.
The evaluation expression acquisition method according to claim 8. In the evaluation expression candidate extraction step,
The evaluation expression candidate extraction unit performs a step of extracting the evaluation expression candidate using the syntax information from a document storage unit that stores a document with syntax information related to words of a document prepared in advance.
In the co-occurrence word extraction step,
The co-occurrence word extraction means performs a step of extracting the co-occurrence word from the document storage means storing the document with syntax information using the syntax information.
The evaluation expression acquisition method according to claim 8 or 9. An evaluation expression acquisition device that acquires an evaluation expression from a document, which is an expression representing an evaluation of the target object, based on an attribute expression that represents the property of the target object,
Attribute expression input means for inputting one attribute expression;
An expression (evaluation expression candidate) that simultaneously appears in the same sentence as the one attribute expression acquired by the attribute expression input unit is related to the appearance position of the word from the document storage unit that stores the word-delimited document prepared in advance. Evaluation expression candidate extraction means for extracting based on information;
Evaluation expression candidate storage means for storing the evaluation expression candidates in the evaluation expression candidate storage means for each sentence;
From the evaluation expression candidate storage means, using the appearance frequency of the evaluation expression candidates, evaluation expression candidate narrowing means for narrowing down the evaluation expression candidates;
For each evaluation expression candidate narrowed down by the evaluation expression candidate narrowing means, words other than the attribute expression that simultaneously appear in the same sentence as the evaluation expression candidate (co-occurrence words), based on the information about the appearance position of the word, Co-occurrence word extraction means for extracting from the document storage means storing the word-delimited document;
Co-occurrence word storage means for storing the set of co-occurrence words in the co-occurrence word storage means for each evaluation expression candidate;
An evaluation expression extracting unit that collectively extracts evaluation expression candidates having the same co-occurrence word in a predetermined number or more from the set of co-occurrence words stored in the co-occurrence word storage unit as an evaluation expression set; ,
An evaluation expression acquiring apparatus characterized by comprising: The evaluation expression candidate extraction means includes:
Means for extracting the evaluation expression candidate using the language information from a document storage means storing a document with language information relating to the word itself of the document prepared in advance;
The co-occurrence word extracting means includes
Means for extracting the co-occurrence word from the document storage means storing the document with language information using the language information;
The evaluation expression acquisition apparatus according to claim 11. The evaluation expression candidate extraction means includes:
Means for extracting the evaluation expression candidate using the syntax information from a document storage means storing a document with syntax information related to words of a document prepared in advance;
The co-occurrence word extracting means includes
Means for extracting a co-occurrence word from the document storage means storing the document with syntax information using the syntax information;
The evaluation expression acquisition apparatus according to claim 11 or 12. Computer
14. An evaluation expression acquisition program that functions as the evaluation expression acquisition device according to claim 11.

Images